Deep Learning Modeling Attack Analysis for Multiple FPGA-based APUF Protection Structures
Ji-Quan Huang, Min Zhu, Bo Liu, Wei Ge · 2018
Arbiter Physical Unclonable Function (APUF) is one of the earliest proposed silicon PUF structures, and its structure and work characteristics have attracted widespread attention and research. However, with the continuous research of PUF, attack techniques for PUF are also emerging. APUF is also vulnerable to modeling attacks due to its own linear delay network structure. In order to increase the anti-modeling attack capability of APUF, various protection structures have been proposed. Deep Learning(DL) modeling attacks have received more and more attention as a new attack method. This paper achieves DL modeling attack on a variety of APUF protection structures based on FPGA platform for the first time. The experimental results show that the DL has strong modeling ability and high prediction accuracy, no lower than 93.56%. Furthermore, there is no need for DL to analyze the protection structure that can accelerate the speed of modeling attacks.